Enhanced Thompson sampling by roulette wheel selection for screening ultralarge combinatorial libraries.

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• The exploration of vast chemical spaces is crucial for identifying novel drug candidates.
• Current methods for selecting compounds from ultralarge virtual libraries face challenges in efficiency, particularly during hit expansion.
• Thompson sampling offers a probabilistic approach to optimize search in reagent space.

Purpose of the Study:

• To address limitations in existing probabilistic search methods for chemical space exploration.
• To develop an improved method for selecting suitable compounds from large virtual libraries.
• To balance greedy search with diversity-driven exploration for more effective hit expansion.

Main Methods:

• Introduction of a roulette wheel selection method.
• Integration of a thermal cycling approach to balance exploration strategies.
• Validation using 109 queries against twenty 1-million-compound libraries with ROCS (Rapid Overlay of Chemical Structures).

Main Results:

• Demonstrated effectiveness of the proposed method in navigating complex chemical spaces.
• Successful application across diverse and large-scale virtual compound libraries.
• Improved compound selection efficiency compared to standard approaches.

Conclusions:

• The combined roulette wheel and thermal cycling method offers a robust strategy for efficient chemical space exploration.
• This approach enhances the identification of promising compounds for drug discovery pipelines.
• The method provides a valuable tool for optimizing hit expansion in virtual screening.